Document 13619595

advertisement
Optimization Methods in Management Science
MIT 15.053
Recitation 2
TAs: Giacomo Nannicini, Ebrahim Nasrabadi
Problem 1
This problem verifies your understanding of the geometry of Linear Programming. The ques­
tions are not trivial, therefore consider carefully your answer.
Consider the feasible region depicted in Figure 2 (note that it extends indefinitely towards
the upper right part of the graph). Here is a proposition that is valid for all linear programs
and will be useful for solving this problem. For any point p, let c(p) be the cost of point p.
Proposition. If point p" is on the line segment joining points p and p"" , then:
• c(p" ) ≥ min{c(p), c(p"" )}, and
• c(p" ) ≤ max{c(p), c(p"" )}.
For example, in Figure 2 we have c(E) ≥ min{c(D), c(F )}, and c(E) ≤ max{c(D), c(F )}.
This follows from simple linear algebra: if p" is on the line segment between p and p"" , then
x2
A
H
B
C
G
D
E
F
Figure 1: Feasible region discussed in Problem 3.
x1
p" = λp + (1 − λ)p"" for some 0 ≤ λ ≤ 1. The cost of a point is a linear fuction and therefore
can be expressed as cTp. Then it is straightforward to verify that cT p" = λcT p + (1 − λ)cT p"" ≥
λ min{c(p), c(p"" )} + (1 − λ) min{c(p), c(p"" )} = min{c(p), c(p"" )}. The proof for the second state­
ment is similar.
Questions. Answer the set of True/False questions below.
(a) F cannot be a unique optimum of the problem.
(b) If C is an optimal solution, D is also optimal.
(c) If A and B are optimal, the problem is unbounded.
(d) If B and F are optimal, G is not optimal.
(e) If no point among B, D and F is optimal, the problem is unbounded.
(f) There exists an objective function such that the problem is infeasible.
(g) If B, D and F are not optimal, the problem is infeasible.
(h) D and F could simultaneously be the only optima of the problem.
(i) If D and G are optimal, there is an infinite number of feasible solutions.
(j) If the problem has a finite optimal objective value, G could be an optimal solution.
(k) There exists an objective function such that H is optimal but A is not.
(l) If H is an optimal solution, there are infinitely many optimal solutions and the limit of the
objective function values is plus or minus infinity.
Problem 2
Consider the feasible region defined by the following constraints:
⎫
x1 − x2 ≥ −1.5 ⎪
⎪
⎬
x1 − 2x2 ≤ 2
4x1 + x2 ≥ 2
⎪
⎪
⎭
x1 , x2 ≥ 0.
(LP2)
(a) Draw the feasible region of (LP2). Does this LP have an optimal solution for all possible
objective functions? Why?
(b) Give an example of:
• An objective function such that both (0, 0) and (0, 1.5) are optimal (in minimization
form)
• An objective function such that only the point (0, 1.5) is optimal (in minimization
form).
• An objective function such that (LP2) is unbounded (in maximization form).
If such an example does not exist, explain why.
2
Problem 3
Consider the following LP:
max
s.t.:
10x1 + 8x2 − 3x3
2x1 + 4x2 − 0.5x3 ≤ 6
−2x1 + 6x2 − 4.5x3 ≤ 4
x1 , x2 ≥ 0
x3
free.
⎫
⎪
⎪
⎪
⎪
⎬
(LP3)
⎪
⎪
⎪
⎪
⎭
(a) Write (LP3) in canonical form. If you have to introduce extra variables, explain what they
stand for. Compute the initial basic feasible solution and write its value for all of the
problem’s variables (regardless of whether they are present in the original formulation or
introduced for the canonical form).
(b) Write the initial simplex tableau and perform two iterations of the simplex algorithm. Is
the basic solution after two iterations optimal? Why?
(c) In (LP3), replace the first constraint 2x1 + 4x2 − 0.5x3 ≤ 6 with −2x1 + 4x2 − 0.5x3 ≤ 6.
Write the initial simplex tableau (note that only one coefficient changes with respect to the
first tableau of Part 2.B). Perform one iteration of the simplex algorithm. What happens
in this case?
Problem 4
Here we review the Simplex Algorithm in more detail, without an Excel spreadsheet to provide
guidance. Be careful when carrying out the calculations.
Consider the following linear program:
max
s.t.:
2x1 + 4x2
0.5x1 − 5x2
x1 + 2x2
x2 + x3
x1 , x2 , x3
≤
≥
≥
≥
12
−2
4
0.
⎫
⎪
⎪
⎪
⎪
⎪
⎪
⎬
⎪
⎪
⎪
⎪
⎪
⎪
⎭
(a) Write the initial simplex tableau. To do so, you will have to transform the problem into
canonical form. What is the initial basic feasible solution? (Hint: at least one of the original
variables is basic)
(b) Perform one iteration of the simplex algorithm, by pivoting in the variable with the largest
reduced cost. Write down the candidate variables for pivoting out, and the corresponding
value of the ratio test. Finally, report the simplex tableau after the first iteration.
(c) Would the simplex algorithm terminate after the first iteration? Why? Can you guess the
optimal objective function value?
3
MIT OpenCourseWare
http://ocw.mit.edu
15.053 Optimization Methods in Management Science
Spring 2013
For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.
Download